Orbifold Zhu theory associated to intertwining operators

Orbifold Zhu theory associated to intertwining

operators

Hiroshi Yamauchi

Graduate School of Mathematics,
University of Tsukuba, Ibaraki 305-8571, Japan
e-mail: hirocci@math.tsukuba.ac.jp

Abstract

We study intertwining operators among the twisted modules for rational VOAs
and show a new modular invariance of the space spanned by the trace functions
associated to intertwining operators. Our result generalizes the results obtained by
Zhu [Z], Dong-Li-Mason [DLM2] and Miyamoto [M3].

1 Introduction

One of the main features of rational vertex operator algebras (VOAs for short) is the
modular invariance. For a rational VOA V, Zhu [Z] proved that the linear space spanned
by the trace functions tr l yio(a)g i( - - ) ~ c ' /24 defined on irreducible V- modules W l is invariant
under the usual action of the modular group SL 2 (Z), where o(a) denotes the grade- keeping
operator (the zero-mode) of a G V. After his work, his result has been generalized in many
directions. Dong, Li and Mason generalized the theory above to the orbifold theory in
[DLM2]. Miyamoto generalized the theory to involve intertwining operators in [M3]. In
this paper, we give a new extension of the Zhu theory which generalizes both the orbifold
case [DLM2] and the intertwining operator case [M3].

Let V be a VOA and G a subgroup of Aut(V). A V-module M is said to be G -stable if
there exists a group representation 7r : G — > End(M) (called the stabilizing automorphism)
such that 7r(A;)Yjw(a, z)v = YM{ka, z)ir(k)v for all a G V, v G M and k G G. Let g, h be
automorphisms on V generating a finite abelian subgroup A of Aut(V). Let (U, <f>) be an
A-stable untwisted ^-module and (W g , ip) an /i-stable g-twisted ^-module, where and
if} are stabilizing automorphisms on U and W g , respectively. We introduce a notion of the
)-twisted intertwining operator of type U x W g — > W g (see Definition 2.2). Let I(-,z)

1

be such an intertwining operator. Then one can consider the trace function associated to
I(-,z) defined as follows:

S T (u, r) = tr Wg z wtu I(u, z)i,{h)q L ^- c ' 2 \ (1.1)

where u G (7 and q = e 2mT . In [DLM2], Dong, Li and Mason studied the trace functions
above in the case U — V, whereas Miyamoto considered them in the case where U is
a general U-module and A = 1 in [M3]. In each case, the space spanned by the trace
functions has a modular invariance property under certain conditions. Therefore, it is
a natural expectation that the linear space spanned by the trace functions (1.1) has a
modular invariance property for general U and A. The purpose of this paper is to prove
that this is true.

This paper contains two main theorems. The first main result is a theorem describing
fusion rules among untwisted and twisted modules in terms of the twisted Frenkel-Zhu's
bimodules (Theorem 3.3). To prove the modular invariance in our case, we need a pre-
liminary result on the fusion rules. So we will extend the theorem describing fusion rules
among untwisted modules which is due to Frenkel-Zhu [FZ] and Li [Li2] to the twisted
case. Using the extended Frenkel-Zhu-Li's theorem, we will prove that the space spanned
by the trace functions (1.1) is invariant under the action of the modular group, which is
the second main theorem.

Let us explain our theorems more precisely. Let V be a VOA and g an automorphism
on V of finite order. Let U be a ^-stable U-module. We will construct an A s (V^)-bimodule
A g (U) as a quotient space of U (see [DLM1] for A g (V)). Using our bimodule A g (U), we
can determine the fusion rules (Theorem 3.3).

Theorem 1. Assume that V is g-rational. Let (U,<f>) be a g-stable V -module and
Wg 1 , W 2 irreducible g-twisted V -modules. Then there is a linear isomorphism between
the linear space of the <fi(g) -twisted intertwining operators of type U x Wg — > W 2 and
Rom Ag(v) (A g (U) ®a 3 {v) W / 9 1 (0),W / 9 2 (0)) 7 where W*(0) denotes the top levels of W l g for
i = 1,2.

Needless to say that, if we consider trivial automorphism, then our construction re-
duces to Frenkel-Zhu and Li's original one (cf. [FZ] [Li2]).

Using the theorem above, we can show a new modular invariance. Let g, h be au-
tomorphisms on V generating a finite abelian subgroup A of Aut(V). Let U be an
A-stable U-module with a stabilizing automorphism <fi : A — > End(C/) and let W g be
an /i-stable g-twisted U-module with a stabilizing automorphism ip{h). Let I(-,z) be a
</>(<?)-twisted intertwining operator of type U x W g — > W g . We say that I(-,z) is h-stable
if tp(h)I(u, z)ip{h)~ l = I((f>(h)u,z) holds for all u G U. Assume that I(-,z) is fa-stable.

2

Then we consider the following trace function:

S\u, t) := tY Wg z wtu I(u, ^(/i)- V (0) ~ c/24 . (1.2)
Our main theorem is the following (Theorem 5.1).

Theorem 2. Assume that V is g-rational. Let (U, <fi) be an A-stable V -module. Assume
that U is C^ ycofinite (see Definition 1^.2). Then the trace functions S t (u,t) defined by
(1.2) converge to holomorphic functions on the upper half plane. Denote by Ci(U; (ft; (g,h))
the linear space spanned by the trace functions S t (-,t) associated to h-stable <ft(g) -twisted
intertwining operators I(-,z) of type U x W g — > W g , where W g runs over h-stable g-

twisted V -modules. For 7 = ^ E SL 2 (Z) with entries satisfying a = d = 1

mod lcm(|</|, \h\), b = mod \g\ and f = mod \h\, define its action on S t (-,t) as
follows:

S 7 | 7 Kt) := Ur + dr k S\u^T),
where u e — {u e U \ L[0]u = ku}. Then 7 defines a linear isomorphism on

C^u-fo (g,h)).

The argument used in the proof of Theorem 2 is almost the same as those in [Z],
[DLM2] and [M3]. But it is worth pointing out the following remarkable difference. In
[Z], [DLM2] and [M3], traces are taken over each irreducible modules. But in our case we
have to handle their conjugate under the automorphism at the same time so that traces
are taken over not only irreducible modules. This is the most difficult point of this paper.

The organization of this paper is as follows. In Section 2 we recall basic definitions
of the twisted theory of VOAs that are needed in later sections. The notion of the <p{g)-
twisted intertwining operators are also given. In Section 3, after introducing the twisted
Frenkel-Zhu's bimodules, we extend Frenkel-Zhu and Li's theorem on fusion rules to the
twisted case. In Section 4 we show the convergence of the trace functions defined as (1.1).
In Section 5 we prove the modular invariance of the space spanned by the trace functions.

2 Basic definition

The most notation are adopted from [DLM2] and [M3]. Let V be a VOA and g an
automorphism of V of finite order \g\. In this paper, we treat only (yf-rational VOAs.

Definition 2.1. A VOA V is called g-rational if all admissible g-twisted V-modules are
completely reducible.

3

Let (Wg,Y g ) be a g-twisted V-module. Take any h G Aut(V). Then one can find a
natural g h = h~ 1 gh-twisted ^-module structure on a vector space W g . Define g h -twisted
vertex operator (Y g ) h on W g by

(Y g ) h (a,z):=Y g (ha,z). (2.1)

Then (W g , (Y g ) h ) has a g A -twisted ^-module structure (cf. [DLM2]). We denote it by
W g o h. In this paper, we consider the case where g and h generate a finite abelian
subgroup in Aut(V). A ^-twisted ^-module W g is called h-stable if W g o h is isomorphic
to W g . That is, there exists a linear isomorphism ip(h) on W g such that

Y g (ha, z)=Mh)Y g (a,z)i/>(h)- 1 .

We call ip{h) a stabilizing automorphism. Note that the stabilizing automorphism tp(h)
is not uniquely determined since non-zero scalar multiplications are allowed.

Let U be a g-stable untwisted K-module admitting a weight space decomposition
U = Q)neNU n+ h and let W g , W g be irreducible p-twisted V- modules admitting weight
space decompositions Wl = © ne j_ N (W / ( !) n+ / l . for i = 1,2, respectively, where we set

y Isl y 1

X s := {v E X \ L(0)v = sv} for a V-module X. One can find the general definition of
intertwining operators among twisted modules in [DLM3] or [X], but we adopt a modified
definition as shown below. Recall the definition of vertex operators on g-twisted V-
modules. Under the action of g, V decomposes as follows:

V = V° (SV 1 ®---®v l9h \

where V r := {a G V | ga = e 2mr ^ 9 ^a}. Let M g be an irreducible admissible (/-twisted
V-module. By Theorem 8.1 (c) of [DLM1], one has a weight space decomposition M g =
® n& ^(M g ) k+n , where A; is a suitable rational number. Setting M 3 g := (B neN (M g ) k+n _j/\ g \
for < j < \g\ — 1, we find that the vertex operators on M g provide the following fusion
rules among \/°-modules V % and Mp

V i x M 3 g = M l g +j , (2.2)

where by abuse of symbols we denote both an integer between and \g \ — 1 and its residue
class mod \g\ by the same letter. To introduce fusion rules similar to (2.2), we assume
that U has the following decomposition

U = U° © U 1 © • • • © U^- 1 (2.3)

as a l /0 -module and vertex operators on U provide the following fusion rules for V°-
modules:

V i x U j = U i+j . (2.4)

4

Based on the decomposition and fusion rules above, we define an intertwining operator
/(•, z) of type U x W 1 — > W 7 " 2 as follows.

Definition 2.2. Let [/ = © ne N^n+/i be a g-stable K-module which has a decomposition
(2.3) with fusion rules (2.4) and let Wl = © ne 1 N (W^) n+ h t , % — 1,2, be irreducible
twisted ^-modules. An intertwining operator of type [/ x Wg — ► W^ 2 is a linear map

/(•, z) : u (E U I— > J(u, 2) = "n*""" 1 G Hom % 2 ) M

neC

with the following properties;
1° for u G U r ,

I(u,z) G Hom(^,lf 9 2 )

where h — h + h± — h 2 \

2° for any I G C, there exists an N G N such that Ui +n w l = for all n> N;

3° L(— l)-derivation : I(L(—l)u,z) — -^-I(u,z);

dz

4° for a E V r and u G U, the following twisted Jacobi identity holds:

Z - 1
"1

r

-i x f zi- z \ f Zi- z \ ^ , , ,
= z 2 6 I — II — I i (Y^a, 2o)«, 22) .

Note that our assumptions (2.3) and (2.4) are equivalent to the existence of a g-
stabilizing automorphism <p(g) on U such that the decomposition (2.3) coincides with the
eigenspace decomposition U r = {u G U\(p(g)u = e 2mr ^ 9 ^u}. Because our definition of
intertwining operators is depend on the choice of the decomposition of U, which owe to
the choice of 4>(g), we call intertwining operators of type U x Wg — > Wg defined as above
" <p{g) -twisted 1 to emphasize the choice of the stabilizing automorphism <f)(g).

The following two properties are direct consequences of the twisted Jacobi identity;

(1) Commutativity: for sufficiently large N G N, we have

(zi - z 2 ) N {Y W 2(a,z 1 )I(u,z 2 ) - I(u,z 2 )Y w i(a,z 1 )} = 0.

(2) Twisted Associativity: for a G V r and w 1 G Wg, let A; be a positive integer such that
z k+ W\Y w i(a,z)w 1 G W£[[z]]. Then

(z + z 2 ) k+ ~\Y W 2(a, z + z 2 )I(u, z^w 1 = (z 2 + z ) k+ T^I (Y v (a, z )u, z 2 ) w 1 .

Conversely, we can prove the twisted Jacobi identity from these two conditions (see [Lil]).

5

3 A generalization of Frenkel-Zhu-Li's theorem

In this section we construct an A 9 (V)-bimodule A g (U) from a (/-stable ^-module U and
show a theorem describing the fusion rules of type (j/^-J in terms of our bimodule A g (U).
These results are basically generalizations of the untwisted case.

3.1 Associative algebra A g (V) and its bimodule A g {U)

Let g be an automorphism of V of finite order \g\ and U a g-stable V-module with
a stabilizing automorphism <p(g). Recall the associative algebra A g (V) introduced in
[DLM1], which is a quotient space of V factored by a subspace O g (V) = (ao g b \ a,b E V).
Let a E V r — {v E V \ gv — e 2mr ^ 9 ^v}. We define a left action of a on U as follows:

(1 + ^wta+|^-l+5,,o

ao g u:=Res z ^$ Y(a,z)u. (3.1)

Using this product we define

O g (U) := ( a o g u | a G V, u e U)

and set

A g (U) := u / (o g (u)nu°),

where U r := {u E U\ 4>{g)u = e 27Tir ^^u}. We define a left action and a right action of
A g {y) on this space. By definition, O g (V) D (B^fl^V 1 " so that A g (V) is isomorphic to a
quotient space of V°. Hence we only need to consider the actions of a E V° on A g (U).
For a E V° we define left and right actions on U as follows:

H _|_ z \™ta

a* g u:=Res z Y(a,z)u, (3.2)

z

n _|_ ^wto-i

u * g a := Res z F(a, (3.3)

When g is trivial, the following theorem holds.
Theorem 3.1. ([FZ]) Ai(U) is an A^V) -bimodule with respect to the actions * 1 .

As an extension of untwisted case we will show

Theorem 3.2. A g (U) is an A g (V) -bimodule with respect to the actions * g .

Proof: We divide the proof into several lemmas. The following lemma will be used
frequently.

6

Lemma 3.1. Let a G V r be homogeneous and let m > n > 0. Then

wta— l+S r ,o + T-T+n

R ^ (1 + ^ + i + Z" ' 9 ' ~ Y{a,z)u G O g {U).

Proof: The proof is similar to that of Lemma 2.1.2. of [Z]. I

We have to show that O g (U) is closed under the actions of V°, the subspace O g (V) acts
trivially on U°/O g (U) fl U° and the associativity of the actions on A g (U). By definition,
we have O g (U) n U° = O(U ) + J^^o yr °g U^~ r \ where O(U ) denotes the kernel of the
(untwisted) Frenkel-Zhu bimodule A(U°) associated to an untwisted y°-module U°. By
Theorem 3.1 we have V° * g 0(U°) C 0(U°) and 0(U°) * g V° C 0(U°). Furthermore, for
a,b E V°, we also have the following by Theorem 3.1.

a* g O(U°) c O(U ),
O(U ) * g ac O(U ),

a *g (b *g U ) - (a *g b) *g C O ( U° ) ,

(U *g a)*gb- U *g (d *g b) C O ( U° ) ,

(a * g U°) * g b-a* g (U° * g b) C O ( U° ) .
Now let / = Xlr^o V r ° g U^~ r . We should show that I is closed under the actions of V°.

Lemma 3.2. Let r ^ 0, a G V* ; 6 G V° and u G £/ l9l ~ r . T/ien both b * g (a o g u) and

(a o g u) * g b are contained in O g (U) fl U° .

Proof: The proof is similar to that of Proposition 2.3 of [DLM1]. i
Next we will show an associative algebra A g (V) acts on A g (U). In other words, we
will show that

(O g (V) n V°) * g U° C O(U ) + I and U° * g (O g (V°) n V°) C O(U ) + I.

By definition, we have O g (V) n V° = O(V ) + J2 r ^o yr °g yl9hr and b Y Theorem 3.1
we have O(V ) * g U° C 0(U°) and U° * g O(V ) C 0(U°). So it remains to show that
(V r oV^- r )* g U° C 0(U°) + I and [7° * g (V r o g V T ~ r ) C O(U ) + 1 for r ^ 0. Let a G V r ,
r ^ 0, 6 G \/ |sl ~ r and u G U°. Then we have

(a Qg b) * g u = Res w > ( . 191 I r(aj_iO, w)w

i=o ^ ? ' w

Res„,Res 2

Wta - 1 + ^ (1 + w )wta+wt6-i

l w
i=0 v 7

{(* - w)^ 1 ^ (a, z)y(6, w)u - (-w + z) i_1 y(6, w)y(a, z)u}

7

= Res^Res,^^ , 1 \ } Y(a, z)Y(b, w)u

w(z — w)

-Res^Res 2 ^^ , 1 + ' Y(b, w)Y(a, z)u

w(—w + z)

Res„,Res 2 ^ 1 uT^l + w) wtb+1 -^Y{a, z)Y{b, w)u

i=0 2

oo x wtb-l+J^p+1

-Res w Res z y (-1)^(1 + 2 ) wta - 1+ W . 1 + W> — W&, iu)y(a, *)u

f ^ ill'-' i ^

■ur

i=0

G 0(E/°) + J.

Similarly, we can show that u * g (a o g b) G 0(U°) + 7.

3.2 Fusion rules

Let V be a rational VOA and let W^, i = 1,2,3, be irreducible V- modules. The following
formula is shown in [FZ] and [Li2]:

dim c [y^f^ = dim c Hom A(v0 (A(W 1 ) ® A (v) W 2 (0), W 3 (0)) .

In this subsection we extend this formula to the twisted modules. Let U be a g-stable V-
module with a stabilizing automorphism 0(g) and let W g , W g be irreducible g- twisted V-
modules. Assume that these modules admit weight space decompositions U = (Bn^nUn+ho
and W l = (B ne i ^{W*) n+ h t , i — 1,2, with some hj G Q, j — 0,1,2, respectively. Set
U(n) := U n+ho and W*{n) := (Wg) n+/lr We may assume that the top levels U(0), Wg(0)
of U, Wg, respectively, are non-trivial. We define deg-u := wtu—h and degu/ := wtw l — hi
for u G U and w % G W g , i — 1,2, respectively.

By definition, a 0(g)-twisted intertwining operator I( • , z) of type U x W^ 1 — >■ W g has
the form

for u <E U r , where h = h + hi — h 2 , and each coefficient U( n ) maps a homogeneous
subspace Wg{m) of W^ 1 into a homogeneous subspace Wg(m + deg-u — n — 1) of W 7 ^.
For homogeneous u G C/°, we denote its weight-keeping operator (the zero mode) by
o 7 (u) := W(degu-i)- We also set o / (-u) = for -u G U r , r ^ and extend it linearly on U.
Similar to the untwisted case we have

Proposition 3.1. o 1 G Hom Ag(v) (A g (U) ® M v) W^{0), W^O)) .

8

Proof: Clearly o 1 is a linear map from U° <8>c Wg(0) to Wg(0). So we only need to
show the following: for a G V° u G U° and w 1 G W^(0),

o / (O 9 (f/)) W 1 = 0; (3.4)
o J (u * a) w 1 = o 1 (u) o(a)w 1 ] (3.5)
and o 1 (a * u) w 1 = o^ajo 1 (u) w 1 . (3.6)

First we consider (3.4). By definition, O g {U) = yr °g U^- r , where v ° °g u ° =

V°oU° (note that o is the untwisted product) and we already know that o 1 (V° o g U°) =
in the untwisted theory (cf. [FZ]). So it remains to show o 1 (j2 r ^o yr °9 £/ M_r ) = 0.
Since this proof is similar to that of Theorem 5.3 of [DLM1], we omit it.

Also (3.5) and (3.6) are known for untwisted intertwining operators and the proof is
similar to the untwisted case. For details, see [FZ]. I

Certain relations between Hornby) [A g {U) ®A g {v) Wg(0),Wg(0)) and the space of
intertwining operators are known.

Proposition 3.2. (Proposition 2.10 of [Li2]) Assume that W l g (i = 1,2) are irreducible.
Then the linear map

* ■■ H; z) G ^ o^g Kom Ag(v) (A g (U) ® Ag{v) W g \0), W 2 (0))

is injective.

When V is rational and g — 1, the linear map it defines an isomorphism (Theorem
2.11 of [Li2]). The following theorem is a generalization of Frenkel-Zhu-Li's theorem.

Theorem 3.3. Let V be a g-rational VOA and (U,4>(g)) a g-stable V -module. Let W l g ,
i = 1,2, be irreducible g-twisted V -modules. Then

) * Hom A9(y) (A g (U) ® Ag(v) W'(0), W 2 (0)) .

Proof: The argument is similar to that in the proof of Theorem 2.11 of [Li2]. i
Remark 3.1. In the theorem above, we don't have to assume the irreducibility of U.

4 The space of trace functions

4.1 The space of 1-point functions on the torus

In the following context, the following notation will be in force. They are taken from
[DLM2].

9

(a) V is a vertex operator algebra.

(b) g, h are automorphisms on V generating a finite abelian subgroup A = (g,h) of
Aut(V).

(c) g has order \g\, h has order \h\ and A has exponent lcm(|g|, \h\).

(d) r(<7, /i) is the subgroup of matrices in SL 2 (Z) such that 6 = mod \g\, c =
mod and a = d = 1 mod lcm(|y|, \h\).

(e) 7i = G C|Im(z) > 0}; the upper half plane.

(f) M(g, h) is the ring of holomorphic modular forms on T(g, h).

(g) U = Q) n mU n +h is an Astable untwisted ^-module 1 with a stabilizing automorphism
(j) : A — > Autc(t/) such that

for all G A

(h) U(g,h) = M(g,h) ® c £/.

(i) E k (r), Pk{n, A,t), Q k (/i,\,T) are defined as follows:

^ ' n=l

(-l) fc ^ A-Hn-j/^-V"-^ g^/M)

and we also set Qo(fJ,, A,r) = —1, where g x denotes e 2 ™, /i = e 2m ^ M , A = e 2ml / N for
integers j, /, M, TV with M, iV > 0, the sign means omit the term n = if (/i, A) = (1, 1)
and Bk(x) are the Bernoulli polynomials defined by the generating function

i=2><4-

k=0

e*

1 We need not assume the irreducibility of U. See also Remark 3.1.

10

For more informations, please refer to [DLM2].
We make it a rule to take

fi r = |/i| r e r(argM)i , where \i = |/i|e (argM)i , —tr < arg/i < n,

for any ^ /i G C and r G Q.

For a VOA (V, Y(- ,z),\,u), one can define another VOA structure on it.

Definition 4.1. (Theorem 4.2.1 of Zhu [Z]) Let (V,Y(- , z), l,u) be a VOA. For each
homogeneous vector a G V, the vertex operator

Y[a,z] := Y{a,e z - l) e zwt{a) = ^ a N^~ n_1 e End(F)[[z,

gives a new VOA structure on V with the same vacuum vector 1 and a new Virasoro
vector uj := uj — ^1.

We denote the Virasoro operators of a new Virasoro vector Co by F[u),z] = J2 n &z
L[n] z~ n ~ 2 . Since L[0] acts on V semisimply, we can decompose V into a direct sum of
eigenspaces for L[0]. Denote a new weight space decomposition of V by V = ©neN^jn],
where Vj„] := {v G V L[0]t> = to }. A V-module U also admits a L[0]-weight space
decomposition. Denote by U[ k ] the L[0]-weight subspace of U with weight fc. For a vector
ti £ {/[J], we define its new weight by wt[u] := k.

Let 0(g, h) be an M(g, /i)-submodule of U(g, h) generated by the following elements,
where a & V satisfies ga = /ia, ha = \a and u G U satisfies 4>{g)u = fjfu, 4>{h)u = X'u for
some fi, /j,', A, A' G C ;

a [0] u, ueU, (//, A) = (1,1), (4.1)

00

a l-2] U + J2( 2k - l ) E ^( T ) ® a [2k-2]U, (/i, A) = (1, 1), (4.2)
fc=2

u, (//,A')^(1,1), (4.3)

00

Qk(n, A, r) <g> a [fc _i]U, (/x, A) ^ (1, 1). (4.4)

fc=0

We define the space of 1-point functions Ci(U; <fi; (g, h)) to be the C-linear space
spanned by functions

S : U(g,h) xH^C

such that

(CI) S(u, t) is holomorphic in r G and for every u <E U(g, h);

11

(C2) S(u, t) is M(g, /i)-linear in the sense that S is C-linear in u and satisfies

S(f®U,T) = f(T)S( U ,T) (4.5)

for / G M(#, /i) and u G U;

(C3) 5(u,r) = if u G 0(#,/i);

(C4) For u G U A — {v E U \ 4>{k)v = v for k G A},

oo

S(L[-2]u, t) = dS(u, t)+J2 E 2l (r)S(L[2l - 2}u, r), (4.6)

1=2

where dS is the operator which is linear in u and satisfies

dS(u, r) = 8 k S(u, r) = ^~^S(u, t) + kE 2 (r)S(u, r) (4.7)

for u G U[k\.

Let f(r) be a holomorphic function on H. We define the weight k action of 7 =

(c 9 e sL2(z) by

(/U 7 )(r):=(cr + rf)- fc /(7r). (4.8)

We also set % := (g a h c ,g b h d ).

The following theorem enables us to prove modular-invariance of trace functions.

Theorem 4.1. ([DLM2J Theorem 5.4, Modular Invariance) For S G Ci(U; 0; (<?, /i)) and
7 = (° ^ G SL 2 (Z) ; de^ne

S| 7 (u, r) := S| fc7 (u, r) = (cr + d)- fc S(n, 7 r) (4.9)
/or w G C/[fc] and extend linearly to U . Then S\^ G Ci(C/; 0; fo) 7 )-

In the rest of this subsection we assume that U satisfies the following finiteness con-
dition which is weaker than the CVcofiniteness.

Definition 4.2. Set C A fi] (U) : = (a[_ 2 ]«, b [0] u \ a G V,b G V A ,u G £/) , where \/ A = { x G
V I /ex = x for all A; G A}. Then [/ is said to be C A yCofinite if dim(U / C A ^(U)) < 00.

Remark 4.1. The Cp ] = C[2,o]-cofinite condition was introduced by Miyamoto in [M3].
It is clear from definitions that C 2 -cofiniteness implies (7[ 2i0 ]-cofiniteness.

Lemma 5.2 of [DLM2] together with suitable modification leads the following.

12

Lemma 4.1. Suppose U is C^yco finite. Then U(g,h)/0(g,h) is a finitely generated
M(g, h) -module.

Lemma 4.2. Suppose U is C^^-co finite. Then for every u G U there exists m G N and
rj(r) G M(g, h), < i < m — 1, such that

m—l

L[-2] m u + r *( r ) ® ^[-2]*M G <3(#, /i). (4.10)

i=0

Proof: Since £/((?, h)/0(g, h) is a finitely generated M(g, /i)-module and M(g, h) is
Noetherian, the submodule generated by {L[— 2] l u \ i > 0} is also a finitely generated
submodule in U(g, h)/0(g, h). So the lemma follows. I

We fix an element S G Ci(U; 0; /i)).

Lemma 4.3. Suppose U is C^^-co finite. Then for every u G U there exists some m G N
and Tiij) G M(g, h) such that

m—l

S(L[-2] m u, r) + J2 n(r)S(L[-2fu, r) = 0. (4.11)

i=0

Proof: Combine Lemma 4.2 together with (C2) and (C3). I

Now S G Ci(U ; <j>; (g, h)) has the same properties as that in [DLM2]. So we can apply
exactly the same arguments as those in [DLM2] (Lemma 6.3-Lemma 6.8 in [DLM2]) and
hence we obtain the following facts: for some p > 0,

v

S(u,T) = Y,QogQx) i Si{»,T), (4.12)

i=0

where

6(i)

S i (u,T)=Y,Q* ij Si j (u,T), (4.13)

oo

E n
a ijn (u)q\o\ (4.14)

n=0

are holomorphic on the upper half-plane, and

\ ih ^ X ij2 (^mod j-Zj (4.15)

for ji ^ j 2 .

Hence we arrive at the following theorem which differs from Theorem 6.5 in [DLM2]
only by changing of C 2 -cofiniteness to Cp O j-cofiniteness.

13

Theorem 4.2. Suppose U is C^^-co finite. For every u G U, the function S(u,r) G
Ci{U;4>; (g,h)) can be expressed in the form (4-12)-(4-15). Moreover, p is bounded inde-
pendently of u.

4.2 /i-conjugate intertwining operators

Let A, U and <fi be as previous. By definition, the stabilizing automorphism <fi gives a
representation of (g) on U. Set V r := {a G V \ ga = e 27nr// ' s 'a} and U r = {u G U \
(f)(g)u = e 27Tir /^u} for < r < \g\ - 1. Then U = U° © U 1 © • • • © U^' 1 and the vertex
operators on [/ give the following fusion rules for \^°-modules:

V i x U j = U i+j .

Namely, the conditions (2.3) and (2.4) are satisfied. Therefore, a stabilizing automorphism
: A — (g,h) — > Autc(f) defines a (g)-grading decomposition on U and hence we can
think of (/>(g)-twisted intertwining operators.

Let Wg, % — 1,2, be irreducible g-twisted ^-modules. Recall the /i-conjugates W % g oh of
W 7 ^ defined by (2.1). By definition, there exist linear maps ipiih) : W l g o — > W 7 ^ (i = 1, 2)
such that

lVi(/ia,z) = V i (/i)yVi 0fc (a,2:)^i(/i) _1 . (4.16)

Assume a 0(g)-twisted intertwining operator /(• , z) of type U x W 7 ^ — > Wg is given. We
can define its /i-conjugate intertwining operator I h (- , z) of type U x IT^ 1 o h — > W 7 ^ o /i as
follows:

J ft ( M , z) := ^(/i)- 1 / z) (/i). (4-17)

Therefore, there exists a linear isomorphism

(W?\ / Wjoh \
\UWl) \U W]oh)

given by an /^-conjugation /(• , z) \— > J h (- , z).

4.3 Fundamental /i-stable sectors

Let M. = {(Wg, Y a )} a( z\ be the set of all inequivalent irreducible g-twisted ^-modules.
Since we have assumed that V is ^-rational, A is a finite set and every irreducible g-twisted
V-module has a weight space decomposition such that all its homogeneous subspaces are
of finite dimension (cf. [DLM1]). We consider a certain equivalent relation on A4. Let
(Wg, Y a ) be an irreducible g-twisted ^-module in Ai. Then an /i-conjugate (Wg, (Y a ) h ^j
is also an irreducible g-twisted K-module and hence there exists a unique (3 G A such

14

that (Wg, (Y a ) h ^ is isomorphic to (W^Y 13 ). Therefore, by defining h(a) = (5 for each
a G A in this way, fa defines a permutation on the index set A and hence on Ai. We set
(W a ,Y a ) oh := (W h ^ a \Y h ^). By definition, there exists unique linear isomorphisms
if) a up to scalar multiplications such that the following commutative diagrams hold for all
a G A:

Wg (a)

Y h ( a \a,z)

w h g (a

>

w a

9

Y a (ha,z)
>

w a .

9

We define an equivalent relation on M. as follows. Define (Wg, Y a ) ~ (Wg 3 , Y 13 ) if both
{Wg, Y a ) and (Wg 3 , Y&) belong to the same orbit under the action of fa, or equivalently
there exists some i > such that (W°,Y a ) o fa* ~ (W^Y 13 ). Take an element (W*,Y a )
in M. and fix it. Assume its orbit {{Wg, Y a ) o fa 1 } (i — 0, 1, . . . , n — 1) is of length n.
For simplicity we denote (Wg,Y a ) o h l by (Wg,Y l ). Then we have linear isomorphisms
i/>i(h) : W 9 o h i+1 -> W s o fa*, i = 0, • • • , n - 1 (g Z/nZ), such that

F*(faa, 2) = ^(fa)F m (a, z)^(fa) -1 . (4.18)

n— 1

Set := ^j-^ and i/>(fa) := V'o(^) © ' ' ' © VVi-i(^)- Then with an automorphism

i=0

^(fa) is the minimal fa-stable g-twisted ^-module in the sense that it contains Wg, i =
0, 1, . . . , n — 1, as submodules with multiplicity one. In this paper, an fa-stable g-twisted
V-module W g constructed as above is of special importance so that we call the modules
constructed as above fundamental h-stable g-twisted V -modules.

4.4 Trace functions

Let M g be a fundamental fa-stable g-twisted V^-module with an fa-stabilizing automor-
phism ip(h). Because V is g-rational, there exists an irreducible g-twisted l^-submodule
W g such that M g = @™~qW 9 o fa*, where n is the minimum positive integer such that
W g o h n is isomorphic to W g . Let /(• , z) be a 0(g)-twisted intertwining operator of type
U x Mg — * Mg. Since M g is fa-stable, an fa-conjugate intertwining operator I h (- ,z) =
■^(fa) _1 /(0(fa) • , z)ip(h) is also of type U x M g — > M g . Hence, fa acts on the space (jf 4 ^ )•
Since we need a nice relation between <f>(h) and ip(h), we adopt the following definition.

Definition 4.3. A 0(g)-twisted intertwining operator /(• , z) of type { V M ^ ) is called
h-stable if tphl( u i z )^h 1 = H4>(h)u, z ) holds for any u E U.

15

The condition that /(•, z) to be fa-stable is equivalent to that /(•, z) is stable under the
action of fa on (jf 4 ^ )• This is why we call "stable". Assume that I(-,z) is fa-stable. The
main subject we study in the rest of this paper is the trace functions defined as follows:

S z (u,q) := z^tr lMg I(u,z)m-V {0) - c/24 . (4.19)

We claim that the trace function (4.19) associated to an fa-stable 0(g)-twisted intertwining
operator /(•, z) belongs to C\(U ; 0; (g, fa)). We may present a rigorous proof here, but it
will be a repetition of some complicated formal calculations which are the same as those
in [Z], [DLM2] and [M3] so that we omit details. For details, please refer to Section 8 of
[DLM2] and Section 3 of [M3]. Here is the consequence.

Theorem 4.3. Let U be an A-stable untwisted V -module with an A-stabilizing automor-
phism <f>, and assume that U is C^yCO finite. Let M g be a fundamental h-stable g-twisted
V -module with an h-stabilizing automorphism ip(h) and let I(-,z) be an h-stable 0(g)-
twisted intertwining operator of type U x M g — > M g . Then the trace function

S T (u, q) := z- tu tT lMg I(u, z^ihy'q 1 ^ 24 , Q = ^

gives an element o/Ci([/;0; (g, fa)).

By definition, a fundamental fa-stable g-twisted V-module M g is a direct sum of
irreducible modules. So an fa-stable 0(g)-twisted intertwining operator I(-,z) of type
U x M g — > Mg is also a direct sum of intertwining operators among irreducible components.
Let W g be an irreducible submodule of M g . It is clear that on W g , only an intertwining
operator of type U x W g o fa — > W g contributes to the trace function (4.19). So we may
assume that /(•, z) is a direct sum of intertwining operators of type U x W g oh %+1 — > W g oh l
with i G N. We call such an fa-stable 0(g)-twisted intertwining operator fundamental.

5 Modular invar iance

In the previous section we showed that the trace function S' / (-,r) associated to an fa-
stable intertwining operator I(-,z) belongs to Ci(U; <p; (g, fa)). We will prove here that
we can choose a basis of the linear space Ci(U; <p; (g, fa)) consisting of trace functions
S r (-,T). Namely, we shall show that the linear space spanned by S t (-,t) coincides with
Ci(U ; 0; (g, fa)), and at the same time we prove a new kind of modular invariance because
the space C\{U] 0; (g, fa)) is invariant under the action of the subgroup T(g,h) of the
modular group SL 2 (Z) by Theorem 4.1.
First, we need the following lemma.

16

Lemma 5.1. ([DLM2] Lemma 9.2) Suppose a G V satisfies ga = /ia, ha = \a with
//, A G C. Then the following hold:

(i) Seen as an element of U[[qr\\, the constant term of a^u — Ylh=i ^ 2k ( T ) a [2fc-i] u is
equal to a* g u — |a[o]W and the constant term of a^ 2 ] u + Y^k=2^^ — l)-^2fc( r ) a [2fc-2]W is
equal to j^a^u + a o g u if /i — A = 1.

(ii) The constant term of YlkLo Qkif 1 , A, q)a,[k-i\u is equal to —a o g u if /i ^ 1, and the
constant term of Ylk=o Qki.Hi X, q) a [k-i]' u > is equal to —a * g u + ih\ a {o] u = 1, A 7^ 1.

Because (L[— l]a) * g u = a o g u and (L[— l]a)[o] = 0, by replacing a by L[— l]a in the
case where /i — 1 and A 7^ 1, we note that —a o g u also appears in the constant term in
that case. Thus, the lemma above insists that almost all generators of O g {U) appear in
the constant terms of elements in 0(g,h).

Let us study the relation between Ci(U;4>; (g,h)) and our trace functions. Take an
arbitrary function S G Ci(g, h). By Theorem 4.2, we know that S is expressed as

p

S(u,r) = £(toggi t)
* — ' w\

i=0

for a fixed p and all u G C/ with each Si satisfying (4.13)-(4.15).
We define a new operation oj* t on U by

* T u := L[-2]u - E 2k {r)L[2k - 2]i

k=l

for u E U. For convenience, we also write

a; * r S(u, r) = 5(0) * T u, r).

In the following, q x means e 2mx .

Lemma 5.2. We have

Sij(a[ ]U,T) = for any a eV A ,u e U,i,j, (5.1)

Sij(u q ,r) = for any u q E 0(g,h),i,j, (5.2)

S , pj (o;* r M,r) = -i-g^-^-S'p j (M,r) for u E U, (5.3)

Si(a>* T ii,T) = t^t < (i + 1)^+1 (u,r) + gi -^—S^u.t) } for u e U, (5.4)
1 rf \ w

and I uj * T — — -q 1 • Sij(u, r) = for N 3> 0. (5.5)

17

Proof: Since 0(g,h) G t/[[g w ]], (5.1) and (5.2) are trivial. Note

1 d 1 1

r9 i

2ni dr \g\ Ja\ dq±_

9

so that from (C4) we have

S{Q * T u, r) = -1 £ U (log ^) 4 1 S t (u, r) + (log q^f -^-S^ r)

|y| i=0 I H JS\

Therefore, we get (5.3) and (5.4). In particular, we find that

oj * T -^-q i — ) Si(u, t) = ^-(i + l)S i+1 (u, t)
\9\ / \9\

and hence (5.5) also holds.

We note that both Si(u,r) and the trace functions S t (u,t) are expressed as a linear
combination of power series of the form q x Yl^Lo a n(u)q n ^ 9 ^. Moreover, both Si and S 1
satisfy the equations (5.1), (5.2) and (5.5). Therefore, we shall investigate the power series
with such properties for a while.

Lemma 5.3. Assume a formal power series T(u,t) = q x J2'^ =0 o: n (u)q n ^ 9 ^ satisfies the
conditions (5.1), (5.2) and (5.5). Then the coefficient of the leading term «o satisfies
the following properties for a G V and u G U such that ga = a, (fi(g)u = u, ha = £a 7
4>{h)u = (u with (,( 6 C :

(i) a (a* g u) = ^~ 1 5^ tl a (u* g a), (5.6)

(a) ao(u') = o for u' e o g {u) + ur i ( 5 - 7 )

(in) a ((co * g -c/24 - X) N v) = for any v E U. (5.8)

Proof: First, we prove (ii). Recall that O g (U) + ^2 r ^ U r is generated by w G U r ,
r ^ and b o g v with b G V r , v G U^~ r , where V r — {x G V \ gx — e 2mr ^ 9 ^x} and
U s = {y G U | (fi(g)y = e 2ms ^ 9 ^y}. Since u> is contained in 0(g,h), we have T(w,t) =
and hence ao( w ) = 0. If r 7^ 0, then by Lemma 5.1 (ii) we get a^(b o g v) = 0. Therefore
we should show a^(b o g v) — for b £ V° and t> G C/°. If hb = b, then by Lemma 5.1 (i)
we have ao(b ° g v + j^b^v ) = 0. On the other hand, by the definition of 0(g, h), we have
cxo(b[o]v) = and hence ao(b o g v) = 0. If hb = eb with e ^ 1, then by Lemma 5.1 (ii) we
obtain a (b o g v) = 0. Thus the assertion (ii) holds.

18

Let a, u, £ and £ be as stated. Then one can show the following equality.

i

"IsT

a * g u — u * g a = a[o]M- (5.9)

Suppose £ = £ = 1. In this case one has a (a[ -\u) = and so that a (a * g u) = a (u * g a).
If £ 7^ 1, then by Lemma 5.1 (ii) the constant term of YlkLo QkO-i £> Q) a [k-i\ u is equal to
— a* g u + Y3|a[o]M (note that in this case a G" V A so we don't know if T(a[o]U, r) = 0). Since
a G V° and T vanishes on YlkLo Qk(l, £; t)o[A;-i] m in this case, we can use the relation
(5.9) to have

a (a * g u) = -j— (ao(° *g u ) ~ a o{u *g a)).

Solving this equation yields (i).

Consider (hi). The constant term of Q * T v is equal to u * g u — \&[o]V by Lemma 5.1.
Since T(Q[ ]V , r) = 0, the leading term of

\g\ T^dq^)

is equal to q x ao((uj* g — c/24 — X)v). Since the operator Co* T ——q^_- will not decrease

the minimal degree of T(v,t), the condition

1 d \ N

W*r- n ?i^ T(v,t)=0

\g\ M dqi I

\s\ /

implies a ((u * T -c/24 - A)^) = 0. Thus (hi) holds. I

By this lemma, the leading term ao of the function T with properties (5.1), (5.2) and
(5.5) can be seen as a linear function on A g (U) with the following /i-twisted symmetric
property: for a G A g (V) and u G A g (U) such that ha = £a, <j>(h)a = (a with £, ( G C,

a (a * g u) = £ _1 %,ia (M * g a).

We shall give an explicit description of such a linear function a .

Proposition 5.1. Let A be a finite- dimensional semisimple associative algebra over C
and let B be an A-bimodule. Let h be an automorphism of an algebra A of finite order.
Assume that a representation <f> : (h) —> Aut(B) of a cyclic group (h) on B is given and
it satisfies

<f>(h){a ■ b) = ha ■ <p(h)b and <j>(h)(b ■ a) = <p(h)b ■ ha (5.10)

19

for a G A and b G B. Let u G Z(A) such that hio — uj. Let F be a linear function on B
with the following properties:

(i) There exists a scalar r G C and N ^> such that

F{{uj - r) N ■ b) = 0. (5.11)

(ii) For ha = aa, <p(h)b = (3b with a, ft G C,

F(a-b) = a- l 5 a p, l F(b-a). (5.12)

Then F can be expressed as

F{b) =Y t C j ti wi I j (b)iP j {h)-\ (5.13)
j

where (W\ iTj), iTj : A — > End(W^ J, ) 7 is an h-stable left A-module on which uj acts as a
scalar r, ipjih) is a linear isomorphism on such that

and Ij is an element o/Hom^(_B <S>a W\ W^) such that

i/jj(h) ■ Ijifyw = Ij {(f>{h)b) ■ ^j{h)w
for a G A, b G B and w G W j .

Proof: By the semisimplicity, A decomposes into a direct sum of simple two-sided
ideals A = ©™ =1 Aj. As an A-bimodule, B is also completely reducible. Furthermore,
the following lemma shows that B decomposes into a direct sum of 0(/i)-invariant A-
subbimodules.

Lemma 5.4. Under the assumption above, for every <p(h) -invariant A-subbimodule B' of
B there exists a <p{h) -invariant A-subbimodule B" such that B = B' © B" .

Proof: Since B is a semisimple A-bimodule, there exists an A-subbimodule B" such
that B = B' ® B". We show that we can choose a 0(/i)-invariant complement B". By the
decomposition of B above, every element b G B can be written uniquely as b = bi + b 2
for some b x G B' and b 2 G B" . Denote the projection map b i— > b 2 with respect to the
decomposition above by a. Clearly, a is an A-homomorphism. Using this projection we
define a linear transformation p on B by

20

Set B'" = pB. We claim that B'" is a 0(/i)-invariant A-subbimodule and B = B' © B'" .
For the 0(/i)-invariance, choose any p(b) G B'" . Then we have

1 \h\-l

<f>(h)p(b) = m ■ w Yl ^ h )~ i(j (^ h y b )

= m E wo* -1 • wo&)

' ' i=0

=p(mb).

Thus 5'" is 0(/i)-invariant. Similarly we can show that the left invariance a- p(b) = p(a-b)
and the right invariance p(b)-a = p(b-a) for all a E A. Therefore, B'" is an A-subbimodule.
Using p 2 = p, one can show that B = B' © B'" . I

By the preceding lemma, we may assume that B doesn't have a proper 0(/i)-invariant
A-subbimodule. Since B is a completely reducible A-bimodule, we can take an irre-
ducible A-subbimodule B x of B. Clearly X^o* 4>(hy Bi is a non-trivial 0(/i)-invariant
A-subbimodule of B, so one must have B = Ylflo 1 0(^)^1 • Take a non-negative integer t
which is a maximal subject to the condition that X^=o < K^)' ? ^i = ©5=o0(^) J '-Si- Then by
induction we see that (p{h) l B 1 C tf^tf^h)' Bi for all Z > and hence B = ©J'^O)^.
Since B\ is an irreducible A-bimodule, only one simple component of A gives a non-trivial
left action on B\. Let us denote such a component by A x . The assumption (5.12) is
equivalent to the fact that

F (a ■ b) = F (b ■ h^a) and F(b) = for <f>(h)b = Xb, A ^ 1. (5.14)

Moreover, one can equivalently replace the condition above by

F(a-b) = F{<P(h)- 1 b-h- 2 a). (5.15)

Let be the unit element of the simple ring A\. For every b\ e Bi, the condition (5.14)
implies that

F(6i) = F(l Al -b 1 )=F (6i • h- l l Al ) = F (6i • U-mJ .

So if i?i is not a non-trivial right h^ 1 Ax-module, then F(Bi) = 0, and by (5.15) we
arrive at F(B) = 0. Therefore, we may assume that B 1 is a (Ai, /i _1 A 1 )-bimodule.
Since h permutes the simple components of A, each irreducible component (f)(h) : 'B 1 of
B is an irreducible (hPAi, /i J ~Mi)-bimodule. So we may assume that h acts transitively
on A as a permutation of its simple components. This means, we may assume that
A = A 1 © hA\ © • • • © h^Ai and h maps /i s-1 Ai onto Ai. In particular, h s induces

21

an automorphism on each simple component h l A\ of A for < % < s. Since <fi{h) 3 B\
is an irreducible (h?Ai, /i J ' -1 Ai)-bimodule, we note that s < t. Furthermore, <f)(h)B =
B forces to be an irreducible (Ai, ftrM^-bimodule so that we obtain s\t. As

B = ®jl} <f>(hy Bi, one can find that the number of isotypical components of B as an
A-bimodule is s. Therefore, B decomposes into a direct sum of its isotypical components
as

s-1 t/s-l

5 = B {k \ where 5 (fc) = (f)(h) si+k B^

k=0 i=0

Hence, <fi(h) induces a cyclic permutation on the set of isotypical components as <f>(h) :
B(k) — > ano \ thus 4>(h) s induces a linear transformation on each isotypical compo-

nent of B.

By definition, B^> is an (A±, ftrM^-bimodule. We twist the right action of A on B^
by h to obtain an Ai-bimodule structure on B^ . Define a new right action of a G A by

b * a := b ■ {h~ l a) for b G £ (0) .

Under the action above, B^ becomes an Ai-bimodule and B^ decomposes into a direct
sum of t/s irreducible Ai-subbimodules as B^ = 0*= o _1 -B O j. Since each irreducible
component Bqj is isomorphic to A\ as an Ai-bimodule, there exist Ai-isomorphisms fj :
A 1 ^ B 0j for < j < t/s — 1. Using these //s, we define linear functions : A x — > C
by

Gj : a G Ai h F (/j(ai)) eC, < j < t/s - 1.
Then for a±, a 2 G Ai, we have

Gj(aia 2 ) = F(f j (a 1 a 2 )) = F (a x ■ fj(a 2 )) = F(f j (a 2 ) ■ h' 1 ^)
= F(fj(a 2 ) * ai) = F(fj(a 2 ■ ai)) = Gj(a 2 ai).

Therefore, the linear functions Gj are trace functions. Let IU be an irreducible left A\-
module. Then Gj can be written as Gj(a) = Gjtr^(a) with some Cj G C and the linear
function F restricted on B j can be described as

F(b j ) = C j ti w fr\b j ) (C,GC) (5.16)

for b G -Boj- To get a desired description of F, we have to extend W to be an A- module.
Based on a left Ai-module structure on W, we construct left /i fc Ai-modules h k o W,
< k < s — 1. As a vector space, we set /i fc o VU = W and we denote the elements
of h k o W formally by h k w, where w G W. We introduce an action of h k Ai on h k o W
by h k a ■ h k w := h k (a ■ w). By this definition, h k o W becomes an irreducible left h k Ai-
module. Set W = ©£ /i fc o W. Then W naturally carries a left A-module structure.

22

Recall that A = (B s k z} Q h k Ai and h s is an automorphism of each simple component of
A. By Skolem-Noether theorem, an automorphism h s of a simple ring Ai is realizable
by an inner automorphism of A\. That is, there exists an element 7 G A\ such that
h s ai = 70i7 _1 for all a x G A x . Then the actions of h s on h k A 1: < k < s — 1, are
given by inner automorphisms defined by h kr y G (/i fc Ai)*. Furthermore, an element 7
induces a linear isomorphism ip of by the following way. For < k < s — 2, we set
*jj(h h w) = h k+1 w and we define ^(h^w) = h°^w for k = s — 1. It is easy to check that
ip is a linear isomorphism on VV. This automorphism is characterized by the following
property:

ip(a ■ w) = ha ■ ip(w), a G A, w G VF.

This means that W is an /i-stable A-module with an /i-stabilizing automorphism ijj.
Next we shall define a linear map I : B^ x W ^ W. For bj G -B j and to e f , we

set

and extend bilinearly on x VF. By this definition, we can show that Iq defines an
element of Rom A (B^ ® A W, W). By (5.16), we note that

F{b) = tiwhib)^- 1

for all b G B<® . Up to now, we have determined an explicit description of F restricted on
B(°\ By (5.15), a description of F on B^ completely determines that on B. Set

W/s-l
I I i=0

Since 0(/i) s is an automorphism on B^ and Jo G Hom^I? 0a W, W), using (5.10) we
can show that I' (b) = ip~ s I (4>(h) s b) 4> s is also an element of Hom^ (B^ 0a W,W).
Furthermore, it follows from (5.15) that F (<p(h) s b) = F(b), and thus trwl'oifyip 1 —
tr^I (b)ifj~ l = F(b) for all b G 5 (0) . Therefore, we have

tr^/ (6)^- 1 = tvwhib)^- 1 = F(b)

for all b G B^°\ It is clear that ip~ s Io (4>(h) s b) tp s = Io(b). Hence, by replacing I by I ,
we may assume

il>- s I {<t>{h) s b)r = h{b). (5.17)

As we did before, by twisting the right action by h k , we can introduce an h k A x -
bimodule structure on each isotypical component B (k \ < k < t/s — 1. Since B^ =

23

4>(h) k B(°\ decomposes into a direct sum of irreducible fa fc Ai-bimodules as

t/s-l

B W = ct>(h) k B 0j .

(5.18)

j=0

On this decomposition, we define I k ■ B^ xW W as follows. For <f>(h) k bj, < k < s—1,
bj G B j and w G W, we set

and extend linearly on B^ k \ It follows from the definition that I k G Hom^ {B^® A W , W).
Furthermore, by (5.15), we see that F(b) = tr l yJ fc (6)^ _1 for all b G By extending

the defining region of I k on the whole of B naturally and setting I :— I + 1± -\ \-I s -i,

we see that I G B.om A (B ® A W, W) and F(b) = ti w I(b)%p~ l for all b G B. Moreover, by
(5.17), tp-T(b)w = I (<f)(h)b) -ip(w) for all b G B and w G W. Therefore I and ip satisfy all
conditions we required. Finally, since uj G Z(A), it is clear that oo acts on W as a scalar,
which should be r by the assumption. I
Now we are ready to apply the preceding proposition to ao by setting A = A g (V),
B = A g (U) and F = a . By the previous proposition and Theorem 3.3, we can construct
intertwining operators I k (-,z) among fundamental fa-stable g-twisted ^-modules such
that the traces of the zero mode on the top modules are given by ao- Furthermore, the
difference between the actions of fa-stabilizing automorphisms on the top levels of fa-stable
^-modules and those of fa-stabilizing automorphisms ip given in the previous proposition
are only scalar multiples. Hence, an fa-stabilizing automorphism ip on a top level gives
rise to an fa-stabilizing automorphism on ^-module. Thus we can obtain fundamental
fa-stable 0(g)-twisted intertwining operators I k (- ,z) such that ao is a linear combination
of the leading terms of S Ik (-, r). Therefore, we have the following statement.

Proposition 5.2. Let a be the coefficient of leading term of the power series T(u, r) =
q x Y^Lo a n(u) g n// ' 3 ' which satisfies (5.1), (5.2) and (5.5). Then there exist fundamental
h-stable g-twisted V -modules (W k ,ip k (h)) with minimum L(0)-weights A + c/24 and fun-
damental h-stable <j>(g) -twisted intertwining operators I k (-,z) of type U x Wg — > Wg such

h 0(fa)%)

w := i) k I (bj)ip k w

that

«o(«) = X) C * tT \ wk ° Ik ( u )Mh) 1 -

k

Thus we have

Proposition 5.3. Suppose U is C^ Q yCofinite. Assume a formal power series T{u,t) =
Q X Yl^=o a ri/\g\( u )Q n ^ g ^ i s given and satisfies (5.1), (5.2) and (5.5). Then there exist fun-
damental h-stable g-twisted V -modules (W k , ^(fa)) and h-stable <f>(g) -twisted intertwining

24

operators ,z) of type U x Wg — * Wg such that T(u,t) can be expressed as a linear
combination of the trace functions S Ik (u, r).

Proof: By Proposition 5.1, there exist fo-stable A g (V)-modu\es (W*(0), ipk(h)), 1 <
k < m, on which u acts as a scalar A + c/24 and linear maps

h e Rom Ag{v) (A g (U) ® Ag{v) W*(0) o h, W^O)), 1 < k < m

such that

m

fyn(u.) = > Ga tr.

By Theorem 3.3, there exist fundamental /i-stable g-twisted V^-modules (Wg, ipk(h)) and
fundamental ^-stable 0(g)-twisted intertwining operators /&(• , z) of type U x Wg — * Wg
such that all the minimum L(0)-weights of Wg are equal to A + c/24 and whose top
levels are isomorphic to Wjj(0) as A 9 (\/)-modules for 1 < k < m. Then the function
T(u,t) — J2j S 1 ' ( u i T ) a l so satisfies the conditions (5.1), (5.2) and (5.5) and the degree
of its leading term is strictly greater than A. If T(u, r) — S Ij (u, r) is not equal to 0,
then we can repeat the same argument on it. But by the rationality of V, there exist only
finitely many inequivalent irreducible ^-modules. So our procedure must stop in finite
steps. Hence, T(u) can be expressed as a linear combination of S Ij (v,ys as desired. I

Proposition 5.4. Si(-, r) = for i > 0.

Proof: Assume p > 1. By Proposition 5.3, S p -i(-,t) can be written as a linear
combination of the trace functions. Since the trace functions satisfy the condition

S\U* T U,T) = ^±-S I (U,T),

we have ^

S p - 1 {u * T u, t) = ——Sp-^u, t).

On the other hand, by (5.4), we have

Sp-xfi * T u, t) = pS p (u, r) + ^-^-Sp-^u, r).

Therefore we get pS p (u, r) = 0, a contradiction. I

Summarizing up to everything, we arrive at the following main theorem. This is a new
extension of Zhu theory which generalize both the orbifold case in [DLM2] and the inter-
twining operators case in [M3].

25

Theorem 5.1. Let V be a g-rational VOA and U an C£ ycofinite A-stable untwisted
V -module with a stabilizing automorphism <fi. Let {(Wg,ipi(h)), . . . , (W™, ip m (h))} be the
set of all inequivalent fundamental h-stable g-twisted V -modules and let {I k j(-,z) I j =
1, . . . ,r k } be a basis of fundamental h-stable <j)(g) -twisted intertwining operators of type
U x Wg — > Wg for each 1 < k < m. Then the space of 1-point functions C\{U ; 0; (g, h))
is spanned by the trace functions

S^(u,t) = tv w ^ u I kj {u,z)^ k {hY l q L ^~^

for 1 < j < r k , 1 < k < m. In particular, the linear space spanned by the trace functions

(s^{-,t) I 1< j <r k ,l<k<m}

is invariant under the modular group T(g, h), where its action is defined by (4.9).

Acknowledgments

The author wishes to thank Professor Masahiko Miyamoto for his insight-
ful advice. He also thanks to Professor Ching Hung Lam for reading this
manuscript and his helpful comments.

References

[Ab] T. Abe, Fusion rules for the free bosonic orbifold vertex operator algebra, J. Algebra
229, 333-374 (2000)

[AM] G. Anderson and G. Moore, Rationality in conformal field theory, Commu. Math. Phys.
117 (1988), 441-450.

[Bo] R.E. Borcherds, Vertex operator algebras, Kac-Moody algebras and the Monster, Proc.
Natl. Acad. Sci. USA. 83 (1986), 3026

[DLM1] C. Dong, H. Li, G. Mason, Twisted representations of vertex operator algebras, Math.
Ann. 310 (1998) 571-600.

[DLM2] C. Dong, H. Li, G. Mason, Modular-invariance of trace functions in orbifold theory,
Commu. Math. Phys. 214, 1-56.

[DLM3] C. Dong, H. Li, G. Mason, Simple currents and extensions of vertex operator algebras,
Commu. Math. Phys. 180, 671-707.

[DM] C. Dong, G. Mason, On Quantum Galois Theory, Duke Math. J. vol.86, No.2 (1997)
305-321.

[FHL] I. Frenkel, Y.-Z. Huang and J. Lepowsky, On axiomatic approaches to vertex operator
algebras and modules, Memoirs Amer. Math. Soc. 104, 1993

26

[FLM] LB. Frenkel, J. Lepowsky, A. Meurman, Vertex Operator Algebras and the Monster,
Academic Press, New York, 1988.

[FZ] LB. Frenkel, Y. Zhu, Vertex operator algebras associated to representation of affine
and Virasoro algebras, Duke Mathematical Journal 66 (1992), 123-168.

[La] S. Lang, Elliptic Functions, Springer- Verlag, 1987.

[Lil] H. Li, Local systems of twisted vertex operators, vertex operator superalgebras and
twisted modules, Cont. Math. 193 (1996) 203-236

[Li2] H. Li, Determining fusion rules by yl(y)-modules and bimodules, J. Algebra 212, 515-
556.

[Li3] H. Li, An analogue of the Horn functor and a generalized nuclear democracy theorem,
Duke Math. J. 93 (1998), 73-114.

[Ml] M. Miyamoto, A modular invariance on the theta functions defined on vertex operator
algebras, Duke Math. J. Vol. 101 (2000), 221-236.

[M2] M. Miyamoto, Modular invariance of trace functions on VOA in many variables,
preprint.

[M3] M. Miyamoto, Intertwining operators and modular invariance, q-alg/0010180.

[X] X. Xu, Intertwining operators for twisted modules of a colored vertex operator super-

algebra, J. Algebra 175, 241-273.

[Z] Y.Zhu, Modular invariance of characters of vertex operator algebras, J. Amer. Math.

Soc. 9 (1996), 237-302.

27